Andreas Beck <[EMAIL PROTECTED]> wrote:
> Ooops - seems like I opened a can of worms here ...
Second try:
Patch as before plus:
Obviosly we should not bail out even on a bogus request with
locks retained:
@@ -843,14 +849,13 @@
individual origin. */
if (inp == NULL) return GGI_EARGINVAL;
+ if (inp->next == inp) return GGI_ENOTFOUND;
if (_gii_threadsafe) {
ggLock(inp->mutex);
ggLock(inp->queue->mutex);
}
- if (inp->next == inp) return GGI_ENOTFOUND;
-
idx = inp;
do {
if ((origin == idx->origin) ||
and we seem to miss to re-free the queue here
@@ -883,7 +888,10 @@
}
inp->next = inp;
inp->prev = inp;
- if (_gii_threadsafe) ggUnlock(inp->mutex);
+ if (_gii_threadsafe) {
+ ggUnlock(inp->queue->mutex);
+ ggUnlock(inp->mutex);
+ }
/* newhand may still be a joined input,
* so we inform the user of that. */
return 1;
The "straight" case seems to handle that right.
Fix works for me. I can now dynamically create and destroy "views".
Shall I commit ? (I hate to fix things the wrong way, so I rather ask one
time too much ...)
CU, Andy
--
Andreas Beck | Email : <[EMAIL PROTECTED]>
